39 Advanced Research Resource: An Introduction to Coding Streams of Language

Any kind of verbal data can be coded. Varying in length, verbal data include the single word responses participants give in questionnaires, the quick posts that participants make in response to news articles, the full texts published in books, articles, and essays—and anything in between. Verbal data may come from conversations that need to be transcribed in order to be analyzed. Or they may come in print form, which may need to be scanned and converted using optical character recognition (OCR). And, increasingly, verbal data come in digital form, har- vested from the web, sent in tweets, or published in digital databases. In most of these cases, verbal data are copious; words come fast and cheap in many contexts. They tell us a lot about what is going on, but we need to work to understand their underlying patterns. This is the work of coding streams of language. Usually when we refer to coding, we are referring to an analytic process guided by a set of procedures—a procedural coding scheme—that tells the analyst how to categorize a segment of verbal data by defining and illustrating the use of each coding category. This is the primary kind of coding we deal with in this book. But we will also introduce readers to two other kinds of coding: automated coding, which uses digital searches to automatically identify members of a coding category, and enumerative coding schemes, which list all of the members of its coding categories. As we shall see in Chapter 4, these three kinds of coding can be used on their own or in combination.

Methodological Approaches to Verbal Data Analysis

Because verbal data are so ubiquitous, many different methodological approaches have been developed to deal with them. Figure 1.1 shows one attempt at displaying complex relationships among these approaches. While coding is an analytic technique used in many fields, it has primarily been developed in the field of communication studies under the term content analysis and in the social sciences, more broadly, under the term qualitative research.

Figure 1.1: Taxonomy of approaches to verbal data analysis.

Traditional quantitative content analysis attempts to remove interpretation from coding. Often used for studies of media coverage, it provides coders with procedures using exact word matches or unambiguous judgments and uses quantification to look at overall patterns. By contrast, qualitative researchers, including those using qualitative content analysis, take an approach that is more interpretative. Many researchers adopt a qualitative approach as part of the process of choosing a CAQDAS (Computer-Aided Qualitative Data Analysis Software) tool such as Nvivo or Atlas.ti. Most though not all qualitative approaches to coding take a code as you go approach, and some, but not all, eschew any kind of quantification. In Coding Streams of Language, we take an interpretive approach to coding; that said, our commitment to being systematic and exploring patterns through numbers places us among the growing number of researchers taking a mixed-methods approach, which we discuss more fully in a later section.

Other methods for verbal data analysis exist that do not use coding. Approaches taken by corpus linguists, for example, focus on analyzing large sets of texts, often using some variety of grammatical or semantic tagging. In Chapter 2 on Designing the Analysis and in Chapter 4 on Coding Data, we suggest ways that one kind of corpus tool, AntConc, can be used to explore and automatically code data.

Finally, emerging methods for data mining have been introduced to deal with large sets of verbal data. Using algorithmic rather than interpretive approaches, many big data approaches have little use for interpretation. But those who use machine learning methods to duplicate human judgment will often begin their work with the kind of coding we pursue.

The Important Role Coding Plays in Many Fields

We come to the coding of verbal data from the allied fields of writing studies and technical communication. No one should be surprised to find these language-intensive fields relying on a method that deals with verbal data. As we noted elsewhere (Geisler, 2018), coding is a key analytic method in writing studies and technical communication, being used in 44% of the research reports published in 2015 and 2016. These reports used a wide range of data. For example, Breuch et al. (2016) coded interview data from hospital patients and their families for recurring themes. Martinez et al. (2015) coded video data for the cognitive activities students used while writing syntheses.

Coding plays an important role in a far wider range of fields than this brief sample of studies might suggest. Any field that deals with humans as social beings, that collects naturally occurring language data or elicits such data from participants, will find a use for coding:

• In applied linguistics, Wyrley (2010) used coding to study communication practices in radiotherapy.
• In education, Stevenson (2013) used coding to study the linguistic strategies used by fifth grade bilingual students in science.
• In engineering education, Richter and Paretti (2009) used coding to analyze how engineering students reacted to multidisciplinary design.
• In information science, Nobarany and Booth (2014) used coding to examine the use of politeness strategies in open peer review.

• In human-computer interaction, Friess (2012), used coding to study the use of personas in software design.
• In legal studies, Jameson, Sohan, and Hodge (2014) used coding to better understand turning points in mediation.
•  In environmental studies,Thompson (2005) used coding to examine the kinds of issues that were discussed in newspaper articles about a proposed off-shore wind power project.
• In public health, Banna et al. (2016) used coding to make a cross-culture comparison of ideas about healthy eating among Chinese and American undergraduate students.
• In operations management, Mugurusi and Bals (2016) use coding to study the stages of an offshoring strategy adopted by a purchasing and supply organization.

When to Code Verbal Data—Or Not

The coding we introduce in Coding Streams of Language is best used when three conditions hold:

1. You are looking for recurrent phenomena within and across streams of language,.
2. You are interested in understanding underlying patterns of doing and meaning in these streams.
3. You and your co-researchers have sufficient intuitions about these streams to place them into appropriate coding categories.

Let’s take a look at these conditions one at a time.

First, coding is a procedure designed to detect recurrent patterns in a stream of language. If you are looking for phenomena that occurs rarely, the procedural coding we recommend in Coding Streams of Language would be more complex than the rewards would justify. For example, if you are looking for the turning point in a conversation, and you expect there to be, at most, one turning point or perhaps none at all in a given stream, you may be better off using a careful close reading to find it. You might still find useful some of the techniques we describe in Chapter 4 for creating an explicit definition for yourself and your readers, but the other procedures described in this book would be more than you need.

Second, the analytic work we recommend in Coding Streams of Language is designed to examine the underlying patterns of meaning and doing, the ways with words of which participants may be largely be unaware. If, however, you are not concerned with the ways specific words and phrases are deployed and responded to, if you only want to identify the places in which certain topics are discussed, then you may only need to use a more simple topical coding (Geisler, 2018; Saldaña, 2016).

Finally, procedural coding, the primary method described in Coding Streams of Language, is designed to guide coders intuitions toward appropriate coding decisions. As we describe more full in Chapter 4, in some situations, no one outside of the context in which a stream was originally produced may have good intuitions about what the language means or how it works. The level of jargon and specialized knowledge may simply prevent outsiders from understanding what is going on from what is being said. If, for example, your verbal stream is in a language you do not understand, you obviously won’t have the intuitions to code it.

But even if you fully understand the language of a verbal stream, you may not have the intuitions to code it appropriately. In this situation, you have two options. One option is to invite an informant, someone who is familiar with the context of production, to work with you as a coder. Another option is to use the enumerative coding, as described in Chapter 4, in which you list all of the possible words or phrases that you include within a coding category. An enumerative coding scheme has the benefit both of being transparent to your readers and of helping them to better understand intuitively what you intend.

To summarize, we invite you to use the procedures in Coding Streams of Language to code verbal data when you are looking for recurrent and underlying patterns in streams of language and about which you or your co-researchers have adequate intuitions.

The Patterns Revealed by Coding

As we discuss in this book, coding can be used to examine three basic kinds of patterns. The simplest pattern is the one-dimensional analysis we describe in Chapter 6, which asks how verbal data is distributed across a set of coding categories, often across a built-in contrast. Banna et al. (2016), for example, used a built-in contrast across Chinese and American undergraduates to notice differences in the ways they thought about healthy eating. Based on these distributional differences, Banna and colleagues recommended different public health strategies be used in these two communities.

Verbal data that have been coded with more than one coding scheme can be looked at multidimensionally, as we introduce in Chapter 7. Jameson et al. (2014), for example, analyzed conversational interactions that occurred during mediation along two distinct dimensions. First, they coded the precipitants leading to turning points in negotiations, points in which the relationship be- tween the disputants seem to change. Second, they coded for negotiation out- come. This allowed Jameson and colleagues to look for relationships between the two dimensions, the kind of precipitants used, and the outcomes of the mediation. Based on the relationships they saw, they suggested ways that mediators could be more helpful.

The third pattern that can be revealed by coding is temporal. As we ac- knowledge in Chapter 8, temporal analysis deserves to be used more often for what it shows us about streams of language. Mugurusi and Bals (2016) use a kind of temporal analysis to show how the dimensions of Centralization, Participation, Formalization, Standardization, and Specialization changed over four phases in the offshoring process. The authors concluded that the offshoring process may be more disjointed and non-linear than current models in operations management would suggest.

Our Core Commitments

We bring to the task of coding streams of language a set of commitments that we’d like to put on the table from the start. They have served as our points of departure for the process and procedures that you will find in the rest of the book. In this section, we make these commitments explicit not so much to argue for them but so that you can judge for yourself.

Commitment to Being Procedural

Coding Streams of Language is fundamentally a procedural guide. That is, it provides you with a set of step-by-step procedures for coding and then analyz- ing verbal data. We anticipate that, as you grow in experience, you will modify, extend, and even discard these procedures. But our intention is to provide you a very clear basis with which to begin.

You will find that most of this procedural knowledge has not been documented elsewhere. Instead, it most often handed down mentor to student during office hours or shared peer to peer in late night sessions. The trouble with these practices is that they tend to keep cultural knowledge about analysis within a closed inner circle. Not only does this seem unfair to us, but it also keeps these procedures out of the light of day. So we put our procedures out there for you to see, use, question, and refine.

Commitment to the Systematicity of Coding

Coding Streams of Language aims to help you produce a systematic analysis. To be systematic means to follow some articulate orderly procedure. It does not mean you have abandoned intuition—more about this later—but it does mean that you have tried as far as possible to create an analysis that can be replicated: that the coding decisions you make today will be the ones that you agree with tomorrow; that the coding decisions your co-researchers make will be more or less the ones that you would make.

The commitment to systematicity lies behind the importance we give to segmenting verbal data in advance of coding it. And, as we introduce in Chapter 3, choosing the right unit for segmentation is key to developing a coding scheme that works. The commitment to systematicity also lies behind our emphasis on reliability. In Chapter 5 we describe how having someone else try to code your data and then comparing it to your own coding is the eye-opening key to developing a good coding scheme.

Commitment to the Design of Analysis

Coding Streams of Language urges you to design your analysis. Verbal data tends to pile up and overwhelm the best of us. Stepping back to consider how you will design your analysis can help you get a handle on what can otherwise be an enormous task.

In Chapter 2, we suggest that you begin with some initial explorations, sharpening your intuitions about what looks interesting. Then we give you some options on sampling your data, using your research questions to pick out a manageable subset of your data for further in-depth analysis. And finally, we recommend that you build your analysis around a built-in contrast, looking not only at data that you think should reveal the phenomenon in which you are interested, but also at data in which you expect the phenomenon to be absent. Sometimes the best way to know what you’re looking for it to see its absence.

Commitment to the Complexity of Language

Coding Streams of Language takes a rhetorical approach to coding. That is, it acknowledges the complexity of language use. It considers not just what language says—that is, its topics—but also what language does. It assumes that language is more than just a vessel for content, more than a series of topics; that it does as well as means.

Acknowledging the complexity of language also requires us to forgo the expectation that any coding scheme can be absolutely unambiguous. Language will always require the interpretive powers of a language user. Coding does not replace the human coder but provides a guide to our intuitions. The role that context plays in developing these intuitions is inescapable. What words and phrases mean in one context might be quite different in another context. Coding depends, however, on the idea that these intuitions can be developed using a full coding scheme as we discuss in Chapter 4.

A Commitment to Mixed Methods

In Coding Streams of Language, we take a mixed-methods approach to the analysis of verbal data. Adapting the terminology introduced by Vogt et al. (2014), the workflow we advocate moves from coding in words to an analysis that combines qualitative (words), quantitative (numbers), and graphic (charts) representations. Like many mixed-methods researchers, we no longer find it useful to see qualitative and quantitative approaches as opposing meth- odologies, but rather prefer to see them as constituting a useful set of tools (Sandelowski et al., 2009).

Nevertheless, our commitment to mixed methods has lead us to adopt the standard of mutual exclusivity for coding. Mutual exclusivity refers to the requirement that each segment of data should be assigned to one and only one code. Mutual exclusivity is often seen as one of the major dividing practices between qualitative and quantitative approaches to coding. Examined more closely, however, we have found that these two analytic traditions are often closer than we might expect because language is inherently multidimensional.

In practical terms, multidimensionality often means that an analyst considering how to code a piece of language often sees multiple ways to code it. This will be true whether one is approaching coding from the perspective of content analysis, in which the goal is to create mutually exclusive categories, or from the perspective of qualitative analysis, in which double coding is not uncommon. Our method for dealing with the tendency to double code is to dimensionalize the data. As we describe in Chapter 4, rather than seeing the inclination to double code as arising from irreconcilable options, we can turn it into an invitation to develop mutually exclusive codes in different dimensions.

Our commitment to mixed methods also keeps us open with respect to research designs. We agree with Vogt et al. (2014), that the choice of analytic methods is not predetermined by the design of your study. Whether you have collected data in the context of a tightly-controlled experimental investigation or as a result of an extended stay in the field, as long as you have verbal data, you can code it and analyze it following the procedures we lay out in this book.

Selected Studies Using Coding Banna, J. C., Gililand, B., Keefe, M., & Zheng, D. (2016). Cross-cultural comparison of perspectives on healthy eating among Chinese and American undergraduate students. BMC Public Health. Retrieved from https://bmcpublichealth.biomed- central.com/articles/10.1186/s12889-016-3680-y

Breuch, L. K., Bakke, A., Thomas-Pollei, K., Mackey, L. E., & Weinert, C. (2016).

Toward audience involvement: Extending audiences of written physician notes in a hospital setting. Written Communication, 33(4), 418-451.

Friess, E. (2012). Personas and decision making in the design process: An ethnographic case study. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, USA, 1209-1218.

Jameson, J. K., Sohan, D., & Hidge, J. (2014). Turning points and conflict transformation in mediation. Negotiation Journal, 30(2), 209-229.

Martínez, I., Mateos, M., Martín, E., & Rijlaarsdam, G. (2015). Learning history by composing synthesis texts: Effects of an instructional programme on learning, reading and writing processes, and text quality. Journal of Writing Research, 7(2), 275-302.

Mugurisi, G., & Bals, L. (2017). A processual analysis of the purchasing and supply organization in transition: The impact of offshoring. Operations Management Research, 10(1-2), 64-83.

Ngai, C. S. B., & Jin., Y. (2016). The effectiveness of crisis communication strategies on Sina Weibo in relation to Chinese publics’ acceptance of these strategies. Journal of Business and Technical Communication, 30(4), 451-494.

Nobarany, S., & Booth, K. S. (2015). Use of politeness strategies in signed open peer review. Journal of the Association for Information Science and Technology, 66(5), 1048-1064.

Stevenson, A. (2013). How fifth grade Latino/a bilingual students use their linguistic resources in the classroom and laboratory during science instruction. Cultural Studies of Science Education, 8(4), 973-989.

Thompson, R. (2005). Reporting offshore wind power: Are newspapers facilitating informed debate? Coastal Management, 33(3), 242-262.

Wyrley, B. (2010). “Talking technical”: Learning how to communicate as a health care professional. South African Linguistics and Applied Language Studies, 28(3), 209-218.

Richter, D. M., & Paretti, M. C. (2009). Identifying barriers to and outcomes of interdisciplinarity in the engineering classroom, European Journal of Engineering Education, 34(1), 29-45.

For Further Reading

Geisler, C. (2004). Analyzing streams of language: Twelve steps to the systematic coding of text, talk and other verbal data. London: Pearson/Longman.